feat(openai): add message_transform hook to OpenAI providers

[programminx-askui]

What & why

Some OpenAI-compatible gateways deviate from the stock OpenAI chat spec (stricter message-ordering, content=null requirements, etc.). Today the only way to adapt the request payload is to monkeypatch the internal _to_openai_messages conversion function — private, process-global, and fragile across upgrades.

This adds a first-class, per-provider hook.

Changes

• OpenAIMessagesApi gains a message_transform parameter. When set, it is applied to the OpenAI-format messages list immediately after conversion and before chat.completions.create(...).
• Threaded through OpenAIVlmProvider and OpenAICompatibleVlmProvider (constructor param + docstrings).
• New public MessageTransform type alias (Callable[[list[dict]], list[dict]]), exported from askui.model_providers.
• Test: tests/unit/models/openai/test_message_transform.py (transform applied vs. pass-through).

Backward compatible — defaults to None, messages sent unchanged.

Usage

from askui.model_providers import OpenAICompatibleVlmProvider

def normalize(messages: list[dict]) -> list[dict]:
    ...  # adapt to the gateway's rules
    return messages

provider = OpenAICompatibleVlmProvider(
    endpoint_url="https://my-gateway/v2/chat/completions",
    model_id="gpt-5.4",
    message_transform=normalize,
)

🤖 Generated with Claude Code